Could you explain more about the data set used in the paper for evaluation? How many features, labels of fault product? And how did you apply the Xception model with the data set?
In this paper, an experiment was conducted using image data of a submersible pump impeller among casting products. The open data set was used for the experiment, and the train data were 3,758 defective images and 2,875 normal image data. And for test data, 453 defective images and 262 normal image data were used.
There are two types of casting data: normal and defective products. Although various defects exist in the abnormal casting data, the label of the defect and the defect segmentation are not performed. In this experiment, data preprocessing was performed by dividing it into normal data and abnormal data.
Using the above data, Xception was used as a deep learning model for product classification. The model structure is composed of three flows: Entry flow, Middle flow, and Exit flow. In entry flow, there are 2 Convolution Layers and 6 Separable Conv Layers.
Middle flow consists of 3 Separable Layers and repeats 8 times.
In Exit flow, 4 Separable Layers are configured. Behind all convolutional and separable layers is batch normalization. And the model was constructed using Maxpooling and ReLu functions.
